Spin exchange in quantum rings and wires in the Wigner-crystal limit

advertisement
INSTITUTE OF PHYSICS PUBLISHING
JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 18 (2006) L7–L13
doi:10.1088/0953-8984/18/1/L02
LETTER TO THE EDITOR
Spin exchange in quantum rings and wires in the
Wigner-crystal limit
Michael M Fogler and Eugene Pivovarov
Department of Physics, University of California San Diego, La Jolla, CA 92093, USA
E-mail: mfogler@ucsd.edu
Received 20 September 2005
Published 9 December 2005
Online at stacks.iop.org/JPhysCM/18/L7
Abstract
We present a controlled method for computing the exchange coupling in
strongly correlated one-dimensional electron systems. It is based on the
asymptotically exact relation between the exchange constant and the paircorrelation function of spinless electrons. Explicit results are obtained for thin
quantum rings with realistic Coulomb interactions, by calculating this function
via a many-body instanton approach.
1. Introduction
Much attention has been devoted to the spin degree of freedom in one-dimensional (1D)
conductors, both of a linear shape (quantum wires [1], carbon nanotubes [2]) and of a circular
one (quantum rings [3–5]). Physical parameters of such systems, e.g., average distance between
the electrons a, their total number N, effective mass m, dielectric constant , etc, can vary
over a broad range or can be tuned experimentally. This creates unique opportunities for
studying the effect of reduced dimensionality and strong Coulomb interactions on quantum
magnetism. A number of new developments have generated a particular interest in the physics
of a 1D Wigner crystal (WC). Unlike the case in higher dimensions, in 1D the crossover to this
strongly correlated regime occurs at easily achievable electron densities [6], rs ≡ a/2aB > 4,
where aB = h̄ 2 /me2 is the effective Bohr radius. Disorder has been the only major obstacle
to realizing the 1D WC experimentally [1]. A promising solution to this problem has been
apparently found, at least, for the case of carbon nanotubes. Very large rs values have been
recently demonstrated in suspended nanotube devices without appreciable intervention of
disorder effects [2]. Because of their finite length, in the desired range rs > 4 these devices
contained only a few electrons, N < 25. Such finite-size systems are traditionally referred to
as Wigner molecules [7]. The progress towards realizing Wigner-crystal (molecule) states in
GaAs quantum wires has also been very encouraging [1]; therefore, one may hope that they
will soon follow suit.
On the theoretical side, the 1D WC is interesting because of a dramatic difference
between the characteristic energy scales for orbital and spin dynamics. This strong spin-charge
0953-8984/06/010007+07$30.00
© 2006 IOP Publishing Ltd
Printed in the UK
L7
L8
Letter to the Editor
(a)
(b)
0.0
τ
9.4
-0.2
tr Ω
uj / a
-0.1
9.2
-0.3
9.0
-0.4
0.0
0.5
1.0
x /a
-0.5
0.0
0.5
1.0
τ
1.5
2.0
Figure 1. The instanton trajectories. (a) Schematic representation for an N = 6 Wigner molecule
on a ring. (b) The trajectories of 1 j 4 electrons for
√ N = 8 (for notations used see the main
text). Inset: function tr Ω(x). The units of τ and Ω are 2a/s0 and its inverse.
(This figure is in colour only in the electronic version)
separation has been recently predicted to cause anomalies in many essential electron properties,
e.g., ballistic conductance [8] of quantum wires and persistent current of quantum rings [7]. In
view of the above, obtaining a reliable estimate of the spin-related energy scales, notably the
exchange coupling J of the nearest-neighbour electrons,is desirable. It has been an outstanding
challenge, though. As depicted in figure 1(a), J is determined by the acts of quantum tunnelling
in which any two such electrons interchange. At rs 1 the corresponding potential barrier
greatly exceeds the kinetic energy of the electron pair, which makes J exponentially small and
difficult to compute numerically [7]. Attempts to derive J analytically (for the nontrivial case
N > 2) were based on the approximation that neglects all degrees of freedom in the problem
except the distance between the two interchanging electrons [8, 9]. We call this the frozen
lattice approximation (FLA). The accuracy of the FLA is unclear because it is not justified
by any small parameter. When a given pair does its exchange, it sets all other electrons in
motion, too (figure 1). To obtain the much needed reliable estimate of J one has to treat the
spin exchange in a Wigner molecule (or a WC) as a truly many-body process. This is done
below in this letter, where we compute J to the leading order in the large parameter rs .
2. Model and results
We assume that electrons are tightly confined in the transverse dimensions on a characteristic
length scale R aB . This allows us to treat the problem as strictly 1D, provided we
replace the Coulomb law by the appropriate effective interaction that goes to a finite value
at distances r R. We adopt a simple model form [10, 11] U (r ) = e2 /(r + R), which
is the simplest expression that correctly captures both short- and long-range behaviour of the
(unscreened) Coulomb interaction for any realistic confining potential and is similar to other
forms used in the literature [12, 13]. For convenience, we focus on the quantum ring geometry
where r = (Na/π)| sin(π x/Na)| is the chord distance and x is the coordinate along the
circumference.
In the Wigner molecule configuration electrons reside at the corners of a regular polygon.
The effective low-energy Hamiltonian of such a state is given by [7]
h̄ 2 2
L +J
S j S j +1 +
n α h̄ωα ,
(1)
H=
2I
α
j
Letter to the Editor
L9
Table 1. Results for Wigner molecules on a ring (finite N ) and for wires (N = ∞).
N
3
4
6
8
∞
∞-FLA
η
κ
2.8009
3.0448
2.7988(2)
3.18(6)
2.7979(2)
3.26(6)
2.7978(2)
3.32(7)
2.7978(2)
3.36(7)
2.8168
2.20
where L is the centre-of-mass angular momentum, S j are electron spins and n α are the
occupation numbers of ‘molecular vibrations’. At large rs , the total moment of inertia I
and the vibrational frequencies ωα are easy to compute because they are close to their classical
values. In order to calculate J , which is more difficult, we first show that the asymptotically
exact relation exists between J and the pair-correlation function (PCF) g(x) of a spin polarized
1D system. For an ultrathin wire, L ≡ ln(aB /R) 1, it is particularly compact:
J = (e2 aB2 /2L)g (0),
rs 1/L.
(2)
By virtue of (2), the calculation of J reduces to an easier task of computing g(x). Using the
instanton method described below we arrive at the final result
κ
π e2
J=
exp(−η 2rs ),
rs 1.
(3)
5/4
(2rs ) L aB
The values of η and κ are given in table 1. They demonstrate that the FLA [8, 9] errs significantly
in κ, by about 50%, but surprisingly little in η, only by 0.7%.
3. Three electrons on a ring
We start with the simplest nontrivial case: N = 3. Let 0 x j < 3a, j = 0, 1, 2, be the
electron angular coordinates. We will compute the exchange coupling J between the j = 0
and the j = 1 electrons. It is convenient to go to new variables: the relative distance of the
pair, x ≡ x 1 − x 0 , and the location of the j = 2 electron with respect to the centre of mass,
X 2 ≡ x 2 − x cm − a. The motion of the centre of mass x cm can be ignored. We arrive at the
problem with one fast (x) and one slow (X 2 ) degree of freedom. (Classically, X 2 = 0.) The
total potential energy Utot (x, X 2 ) = U (x) + U [(3/2)(X 2 + a) − x/2] + U [(3/2)(X 2 + a) + x/2]
has two global minima in the fundamental domain |x| < 3/2a, at x = ±a, X 2 = 0. They
give rise to the two lowest-energy multiplets: the spin-singlet ground state S0 + S1 = 0 with an
orbital wavefunction s (x, X 2 ) and the triplet with a wavefunction t . Their energy splitting
is the desired exchange coupling J . It is given by the formula [14, 15]
J = (2h̄ 2 /µ) d X 2 1 ∂x 1 |x=0 ,
(4)
where the (normalized to unity) ‘single-well’ wavefunction 1 (x, X 2 ) is the ground state
of the Hamiltonian with a modified potential Utot → U1 ≡ Utot (max{x, 0}, X 2 ) and
µ = m/2. Equation (4) is valid to order O(J 2 ) [14]; with the same accuracy, the singlet
and triplet wavefunctions are symmetric and antisymmetric
combinations of the single-well
√
wavefunctions, s,t = [1 (x, X 2 ) ± 1 (−x, X 2 )]/ 2.
Let us discuss the form of 1 (x, X 2 ). Near its maximum at x = a, X 2 = 0, it is a simple
Gaussian in both variables, characterized by an amplitude l of the zero-point motion in x and
a frequency (a) of the zero-point oscillations in X 2 . Away from its maximum 1 rapidly
decays at |X 2 | l a. This justifies the following Gaussian approximation1 in the entire
fundamental domain of x:
(5)
1 = φ(x) exp[−(M/2h̄)
(x)X 22 ],
1
The Gaussian ansatz has been used previously for computing spin exchange in 3 He crystals [16].
L10
Letter to the Editor
g
I
0
II
III
aB
a
x
Figure 2. The PCF of a spin-polarized system (schematically). Regions I, II and III are described
in the main text.
where M = 3µ. It is important that at x a, where the tunnelling barrier is large, is a
slow function of x. Hence, if g(x) denotes the PCF of a spin-polarized molecule,
N
−1
d X j 2t (x, X 2 , . . . , X N −1 ),
|x| < 3a/2,
(6)
g(x) ≡ 2
j =2
then (4) immediately entails J = (h̄ 2 /4µ)[φ(0)/φ (0)]g (0). Anticipating the discussion
below, (6) is written for an arbitrary N > 2, with the notation X j ≡ x j −x cm +(N −1−2 j )(a/2)
3a/2
being used; the PCF is normalized as appropriate in the WC limit, 0 g(x) dx = 1.
The equations on φ(x) and (x) are obtained by substituting (5) into the Schrödinger
equation and neglecting terms small in l/a. This results in the dependence of g(x) on x sketched
in figure 2. Near its x = a maximum (region III) g(x) is a Gaussian of width l. In region II the
quasiclassical approximation applies. Finally, in an ultrathin wire, L 1, there is also region
I, x aB , where the quasiclassical approximation breaks down. Fortunately, the equations
on φ(x) and (x) can be simplified there, as (x) (0) and Utot (x) U (x) + 2U (3a/2).
Similar to [10], this leads to φ(0)/φ (0) aB /L, which, combined with the expression for J ,
yields equations (2) and (3), with η and κ given by
1/2
a
dx a
η=2
U
(x)
,
(7)
tot
a e2
0
3/4
25/4 ξ(0) (a) a 3 U (a)
.
(8)
κ= √ e
(0) e2 tot
π
Thus, for the N = 3 case we were able to reduce the original complicated three-body
eigenvalue problem to routine operations of solving an ordinary differential equation on (x)
and taking two quadratures. The resultant η and κ are listed in table 1. In comparison [8], the
FLA underestimates κ by about 50%. It gets η correctly but only for N = 3; see more below.
One important comment is in order. The antisymmetry of the total fermion wavefunction
imposes certain selection rules [17] for the allowed values of L (see (1)) at a given total spin S.
The lowest-energy L eigenstates for the two possible S values in the N = 3 system, S = 1/2
and 3/2, are |L| = 1 and 0, respectively. Since J h̄ 2 /I at large rs , the ground state of the
system is the L = 0 spin quartet [7, 11].
4. N > 3 electrons on a ring
In a system of more than three electrons, the single-well function 1 (x, X) can be
sought in the form similar to (5), but with the argument of the exponential replaced by
Letter to the Editor
L11
√
−1/2
(−1/2h̄)(X† M1/2 )Ω(x)(M1/2 X), where Mi j = m −1/2 [δi j − (1 − 2/N )/(N − 2)]. In
the language of quantum tunnelling theory, Ω(x) is a matrix that controls Gaussian fluctuations
X = X−X∗ around the instanton trajectory X∗ (x), where X = (X 2 , . . . , X N −1 )T . Switching
to the usual parametrization of the instanton by an ‘imaginary time’ τ , we seek x(τ ) and X∗ (τ )
that minimize the action
∞
dτ µ
1
2
†
(∂τ x) + (∂τ X) M∂τ X + Utot ,
(9)
SN =
h̄ 2
2
0
subject to the boundary conditions x(0) = 0, x(∞) = a and X(∞) = 0. Henceforth Utot is
always meant to be evaluated on the instanton trajectory and Utot stands for the difference
of its values at a given τ and at τ = ∞. Repeating the steps of the derivation for the N = 3
case, we derive the following equations on φ(x) and Ω(x):
∂τ Ω = Ω2 (τ ) − ω 2 (τ ),
{(h̄ 2 /2µ)∂x2 − Utot (x) − (h̄/2) tr Ω(x) + E}φ(x) = 0,
(10)
where ω is a positive-definite matrix such that ω 2 = M−1/2 ΞM−1/2 and Ξ is the matrix
of the second derivatives i j = ∂ X i ∂ X j Utot . The equations are mutually consistent if
E = Utot (a) + (h̄/2)[tr ω (a) + ω0 ], ω0 ≡ h̄/µl 2 and Ω(a) = ω (a).
The PCF g(x) in the quasiclassical region can be written in terms of the tunnelling action (9)
and the appropriate prefactor as follows:
a 1 (a) h̄ω0 1/2 ξ(x)−2S N (x)
g(x) = 2
e
,
(11)
l 2π (x) U (x)
a
ω0 + tr Ω(a) − tr Ω(y)
1
dy
−
ξ(x) =
.
(12)
[(2/µ)Utot (y)]1/2
a−y
x
Here the action SN is defined to be the value of√the integral in (9) when its lower limit is
replaced by τ = τ (x). For η we find η = 2SN / 2rs , while κ is given by equation (8) after
the replacement → det Ω.
5. Calculation of the instanton
A few properties of the instanton follow from general considerations. The dimensional analysis
√
of action (9) yields SN ∝ rs , so that η is indeed just a constant. Also, from the symmetry
of the problem, X N +1− j (τ ) = −X j (τ ). Thus, in the special case of N = 3, the instanton
trajectory is trivial: X 2 ≡ 0, i.e., the j = 2 electron does not move. This is why we were
able to compute S3 in a closed form. For N > 3 the situation is quite different: all electrons
(except j = (N +1)/2 for odd N) do move. In order to investigate how important is the motion
of electrons distant from the j = 0, 1 pair, let us consider the N = ∞ (quantum wire) case,
where the far-field effects are the largest. If the X j were small, we could expand Utot in (9)
to the second order in X j to obtain the harmonic action
1m
dk
dω
|u kω |2 [ω2 + ω2p (k)],
Sh =
(13)
2 h̄
2π
2π
where u kω is the Fourier transform of electron displacement u j (τ ) ≡ x j − x 0j from the
classical equilibrium position x 0j ≡ ( j − 1/2)a, j ∈ Z, ω p (k) s0 k ln1/2 (4.15/ka)
is the plasmon dispersion in the 1D WC and s0 ≡ (e2 /µa)1/2 . Minimization of Sh
with the specified boundary conditions yields u j (τ ) ∝ vx 0j /[(x 0j )2 + v 2 τ 2 ], where v (s0 /2) ln{[(x 0j )2 + s02 τ 2 ]/a 2 }. Substituting this formula into harmonic action (13), we find that
the contributions of distant electrons to Sh rapidly decay with | j |. Thus, a fast convergence of
L12
Letter to the Editor
η to its thermodynamic limit is expected as N increases. Encouraged by this conclusion, we
undertook a direct numerical minimization of S for the set of N listed in table 1 using standard
algorithms of the popular software package MATLAB. The optimal trajectories that we found
for the case of N = 8 are shown in figure 1(b). As one can see, electron displacements
reach some finite fractions of a at τ = 0. This collective electron motion lowers the effective
tunnelling barrier and causes η to drop below its FLA value, although only by 0.7%; see table 1.
Let us now discuss the prefactor κ. In the inset of figure 1(b) we plot tr Ω(x) computed by
solving (10) numerically. To reduce the calculational burden, the matrix ω and the potential
energy Utot in this equation were evaluated on the FLA trajectory X(τ ) = 0 instead of the
true instanton trajectory shown in figure 1(b). The error in κ incurred thereby is ∼2% [19].
In comparison, the FLA, where tr Ω(x) = const, yields κ about 50% smaller than the correct
result, similar to N = 3.
A straightforward interpolation of our finite-N results from table 1 to larger number of
electrons indicates that the changes in η and κ from N = 8 to N = ∞ are smaller than the
accuracy of our numerical procedure.
6. Relation to current experiments
For carbon nanotube quantum dots [2], where the WC limit has apparently been realized, our
formula (3) gives J ∼ 1 K at rs = 4, which should be verifiable experimentally. Unfortunately,
the lowest measurement temperature was 0.3 K; therefore, the exchange correlations may
have been washed out. We hope that our predictions can be checked in the next round of
experiments. Energy-level spectroscopy of quantum rings [3–5] is another promising area
where our results may apply. In longer 1D wires, J determines the velocity vσ = (π/2)J a/h̄
of spin excitations, which can be measured by tunnelling [1], photoemission [18], or deduced
from the enhancement of the spin susceptibility and electron specific heat [10]. Our result for
vσ reads (cf table 1)
vσ /vF = 5.67(π/L)rs3/4e−η
√
2rs
,
η = 2.7978(2),
(14)
where vF = (π/2)(h̄/ma) is the Fermi velocity.
Being asymptotically exact in the rs → ∞ limit, equations (3) and (14) are most accurate
at large rs . Additional arguments [19] suggest that at rs = 3–4, where the Wigner molecule just
forms, the accuracy of these equations is better than 50%. In principle, even higher accuracy
at such rs can be achieved if g in equation (2) is computed by the quantum Monte Carlo
technique [20]. This is worth a separate investigation.
This work is supported by the A P Sloan and the C & W Hellman Foundations.
Note added. After the completion of this work, we learned that Klironomos et al [21] independently computed
η = 2.798 05(5), but not the prefactor κ. These authors also considered a correction to η due to a finite radius of the
wire R. We can show that as R increases the ratio π/L in (14) is replaced by a more complicated expression that
tends to unity at R > aB .
References
[1] Auslaender O M, Steinberg H, Yacoby A, Tserkovnyak Y, Halperin B I, Baldwin K W, Pfeiffer L N and
West K W 2005 Science 308 88 and references therein
Field S B, Kastner M A, Meirav U, Scott-Thomas J H F, Antoniadis D A, Smith H I and Wind S J 1990 Phys.
Rev. B 42 3523
[2] Jarillo-Herrero P, Sapmaz S, Dekker C, Kouwenhoven L P and van der Zant H S J 2004 Nature 429 389
[3] Lorke A, Luyken R J, Govorov A O, Kotthaus J P, Garcia J M and Petroff P M 2000 Phys. Rev. Lett. 84 2223
Letter to the Editor
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
L13
Warburton R J, Schäflein C, Haft D, Bickel F, Lorke A, Karrai K, Garcia J M, Schoenfeld W and
Petroff P M 2000 Nature 405 926
Fuhrer A, Ihn T, Ensslin K, Wegscheider W and Bichler M 2004 Phys. Rev. Lett. 93 176803
Bayer M, Korkusinski M, Hawrylak P, Gutbrod T, Michel M and Forchel A 2003 Phys. Rev. Lett. 90 186801
Egger R, Häusler W, Mak C H and Grabert H 1999 Phys. Rev. Lett. 82 3320
Reimann S M and Manninen M 2002 Rev. Mod. Phys. 74 1283
Viefers S, Koskinen P, Singha Deo P and Manninen M 2004 Physica E 21 1
Matveev K A 2004 Phys. Rev. B 70 245319
Häusler W 1996 Z. Phys. B 99 551
Fogler M M 2005 Phys. Rev. B 71 161304(R)
Usukura J, Saiga Y and Hirashima D S 2005 J. Phys. Soc. Japan 74 1231 and references therein
Friesen W I and Bergerson B 1980 J. Phys. C: Solid State Phys. 13 6627
Szafran B, Peeters F M, Bednarek S, Chwiej T and Adamowski J 2004 Phys. Rev. B 70 035401
Herring C 1964 Phys. Rev. 134 A362
Landau L D and Lifshitz E M 1977 Quantum Mechanics (Oxford: Pergamon) section 81
Roger M, Hetherington J H and Delrieu J M 1983 Rev. Mod. Phys. 55 1
Maksym P A 1996 Phys. Rev. B 53 10871
Koskinen P, Koskinen M and Manninen M 2002 Eur. Phys. J. B 28 483
Claessen R, Sing M, Schwingenschlögl U, Blaha P, Dressel M and Jacobsen C S 2002 Phys. Rev. Lett. 88 096402
Fogler M M and Pivovarov E 2005 Preprint cond-mat/0510294
Bernu B, Candido L and Ceperley D M 2001 Phys. Rev. Lett. 86 870
Klironomos A D, Ramazashvili R R and Matveev K A 2005 Preprint cond-mat/0504118
Download